Executive Summary
A critical vulnerability, CVE-2026-83493, with a CVSS score of 9.1, was discovered in Red Hat OpenShift AI. This flaw allows a remote attacker to exploit the pull_request_target workflow by submitting a specially crafted pull request, potentially leading to unauthorized access to sensitive secrets and control over the repository.
Technical Analysis
The vulnerability is classified as an authorization bypass issue (CWE-829). The attack vector involves a remote attacker submitting a specially crafted pull request to the pull_request_target workflow. This workflow processes untrusted code from the pull request author while operating with the base repository's sensitive secrets, such as the SNYK_TOKEN and its default write token. The root cause of this vulnerability is the improper handling of untrusted input in the workflow.
How It Gets Exploited
An unauthenticated remote attacker could submit a specially crafted pull request to the pull_request_target workflow. When processed, this pull request could lead to the exfiltration of sensitive secrets, such as the SNYK_TOKEN and the default write token. With these secrets, an attacker could gain unauthorized access to the Snyk integration and potentially take control of the repository.
Impact Assessment
Red Hat OpenShift AI (RHOAI) is affected by this vulnerability. An attacker could achieve unauthorized access to sensitive secrets, leading to potential control over the repository and the Snyk integration. The CVSS score of 9.1 indicates a critical severity level, with high impacts on confidentiality and integrity.
Recommended Actions
Update Red Hat OpenShift AI to the latest version available. Specifically, ensure that the rhoai/odh-training-cuda128-torch29-py312-rhel9 package is updated. Block any suspicious pull requests to the pull_request_target workflow, and monitor for unusual activity related to the Snyk integration and repository access.
Sources
- National Vulnerability Database (NVD)
- Red Hat Security Advisory
- Bugzilla Red Hat